A probabilistic approach to calculating probability of achieving target pharmacological effect levels for a novel orally administered antibacterial chemotherapeutic
Raman Sharma, Andrew Skingsley, Pablo Gamallo, Joel Lelievre, Rob Bates, Chao Chen
GlaxoSmithKline PLC
Objectives: To apply a new probabilistic PK-PD (Pharmacokinetics-Pharmacodynamics) framework to predict the probability of achieving target pharmacological effect levels of a novel orally administered antibacterial chemotherapeutic (molecule X) in a Phase 2a study for a potentially life-threatening bacterial infection.
Methods: Phase 1a pharmacokinetic data for molecule X, and in-vitro MIC data were leveraged, alongside multi-source Phase 2a PK and PD data for the dose-fractionation of an in-class intravenously administered benchmark antibacterial drug, in order to determine the probability of achieving the PKPD indices that correspond to target pharmacological effect levels at the currently planned dose of molecule X.
Population PK analysis was performed for molecule X and the benchmark antibacterial using a parametric nonlinear mixed effects modelling approach. For the benchmark antibacterial, post-hoc individual PK profiles were used to calculate individual level PKPD indices using a typical value of MIC (1 ug/mL). In order to determine the most relevant PKPD driver for the benchmark antibacterial drug, correlation of mean PKPD index versus mean effect per dose fractionated treatment arm was examined for PKPD indices AUC(0-24):MIC, %T > MIC and Cmax(dose01):MIC.
Probability distributions for MICs were formally fitted for molecule X and the benchmark antibacterial drug using the available proprietary and literature data. Normal, log-normal, gamma and logistic candidate distributions were fitted using a maximum likelihood approach and assessed using Bayesian information criteria (BIC). Population PK profiles and MIC distributions were sampled and subsequently combined to calculate distributions of PKPD indices for molecule X and the benchmark antibacterial drug. A novel probabilistic approach which takes account of the probability distributions of PK and in-vitro MIC values for molecule X and the benchmark drug, was subsequently employed to calculate probability of molecule X reaching PKPD index values that correspond to target pharmacological effect values for the benchmark drug regimens. To our knowledge this is the first application of this approach to PKPD modelling of anti-infectives.
Results: A two-compartment linear mammillary model with first-order absorption was found to best fit the PK data for both molecule X and benchmark antibacterial drug. Of the three PKPD indices examined, AUC(0-24h):MIC was found to have the highest correlation (R2 = 0.79, p=0.02) for the benchmark antibacterial with %T > MIC (R2 = 0.39, p=0.2) and Cmax(dose01):MIC (R2 = 0.30, p=0.3) showing significantly less correlation. A lognormal distribution (μ = 1.28 ug/mL, σ= 0.54 ug/mL) and logistic distribution (μ = 0.88 ug/mL, σ= 0.30 ug/mL) were found to best fit (ΔBIC > 2) the MIC data for molecule X and benchmark antibacterial, respectively. Taking account of the probability distributions of the PK and PD for both molecule X and the benchmark antibacterial drug, the probabilities of molecule X reaching PKPD indices – including AUC(0-24h) – for the minimally acceptable pharmacological effect level observed in benchmark regimens were found to be all < 10% at the currently planned regimen.
Conclusions: This work highlights the importance of benchmarking pharmacology, leveraging multi-source data, and scrutinizing relative informativeness of multiple PKPD indices when performing model informed anti-infective drug development. Moreover, we show the utility of a new probabilistic approach that takes account the probability distributions of PK and MIC values of the drug of interest and benchmark drug to assess the probability of achieving benchmark PKPD index values, thus allowing the variability and likelihood in benchmark PKPD index to be fully taken into account.